Make sure rdx is zero on return to benefit gcd_22's private calls.

Make the gcd_11 files more similar in register use.

4 files changed, 73 insertions, 68 deletions

diff --git a/mpn/x86_64/bd2/gcd_11.asm b/mpn/x86_64/bd2/gcd_11.asm
index 6260b2536..edf8f4c28 100644
--- a/mpn/x86_64/bd2/gcd_11.asm
+++ b/mpn/x86_64/bd2/gcd_11.asm
@@ -39,13 +39,13 @@ C	     cycles/bit (approx)
 C AMD K8,K9	 -
 C AMD K10	 -
 C AMD bd1	 -
-C AMD bd2	 3.27 *
+C AMD bd2	 3.27  *
 C AMD bd3	 ?
 C AMD bd4	 3.79
 C AMD bt1	 -
-C AMD bt2	 3.64 *
-C AMD zn1	 3.25
-C AMD zn2	 3.50
+C AMD bt2	 3.64  *
+C AMD zn1	 3.25  *
+C AMD zn2	 3.25  *
 C Intel P4	 -
 C Intel CNR	 -
 C Intel PNR	 -
@@ -73,21 +73,24 @@ ASM_START()
 	ALIGN(16)
 PROLOGUE(mpn_gcd_11)
 	FUNC_ENTRY(2)
-	mov	v0, %rax	C
-	sub	u0, v0		C
-	jz	L(end)		C
+	mov	v0, %rdx
+	sub	u0, %rdx
+	jz	L(end)
 
-	ALIGN(16)		C
-L(top):	rep;bsf	v0, %rcx	C tzcnt!
-	mov	u0, %r9		C
-	sub	%rax, u0	C u - v
-	cmovc	v0, u0		C u = |u - v|
-	cmovc	%r9, %rax	C v = min(u,v)
-	shr	R8(%rcx), u0	C
-	mov	%rax, v0	C
-	sub	u0, v0		C v - u
-	jnz	L(top)		C
+	ALIGN(16)
+L(top):	rep;bsf	%rdx, %rcx		C tzcnt!
+	mov	u0, %rax
+	sub	v0, u0			C u - v
+	cmovc	%rdx, u0		C u = |u - v|
+	cmovc	%rax, v0		C v = min(u,v)
+	shr	R8(%rcx), u0
+	mov	v0, %rdx
+	sub	u0, %rdx		C v - u
+	jnz	L(top)
 
-L(end):	FUNC_EXIT()
+L(end):	mov	v0, %rax
+	C rax = result
+	C rdx = 0 for the benefit of internal gcd_22 call
+	FUNC_EXIT()
 	ret
 EPILOGUE()
diff --git a/mpn/x86_64/bd4/gcd_11.asm b/mpn/x86_64/bd4/gcd_11.asm
index d34fb4f2a..4c4cc2dfe 100644
--- a/mpn/x86_64/bd4/gcd_11.asm
+++ b/mpn/x86_64/bd4/gcd_11.asm
@@ -41,11 +41,11 @@ C AMD K10	 -
 C AMD bd1	 -
 C AMD bd2	 -
 C AMD bd3	 -
-C AMD bd4	 2.86  *
+C AMD bd4	 4.0   *
 C AMD bt1	 -
 C AMD bt2	 -
-C AMD zn1	 2.66  *
-C AMD zn2	 3.48
+C AMD zn1	 3.25  *
+C AMD zn2	 3.50
 C Intel P4	 -
 C Intel CNR	 -
 C Intel PNR	 -
@@ -73,22 +73,24 @@ ASM_START()
 	ALIGN(16)
 PROLOGUE(mpn_gcd_11)
 	FUNC_ENTRY(2)
-	mov	v0, %rax	C
-	sub	u0, v0		C
-	jz	L(end)		C
-	mov	u0, %r9
+	mov	u0, %rax
+	mov	v0, %rdx
+	sub	u0, %rdx		C v - u
+	jz	L(end)
 
-	ALIGN(16)		C
-L(top):	rep;bsf	v0, %rcx	C
-	sub	%rax, u0	C u - v
-	cmovc	v0, u0		C u = |u - v|
-	cmovc	%r9, %rax	C v = min(u,v)
-	shrx(	%rcx, u0, %r9)	C
-	shrx(	%rcx, u0, u0)	C
-	mov	%rax, v0	C
-	sub	u0, v0		C v - u
-	jnz	L(top)		C
+	ALIGN(16)
+L(top):	rep;bsf	%rdx, %rcx		C tzcnt!
+	sub	v0, u0			C u - v
+	cmovc	%rdx, u0		C u = |u - v|
+	cmovc	%rax, v0		C v = min(u,v)
+	shrx(	%rcx, u0, %rax)
+	shrx(	%rcx, u0, u0)
+	mov	v0, %rdx
+	sub	%rax, %rdx		C v - u
+	jnz	L(top)
 
-L(end):	FUNC_EXIT()
+L(end):	C rax = result
+	C rdx = 0 for the benefit of internal gcd_22 call
+	FUNC_EXIT()
 	ret
 EPILOGUE()
diff --git a/mpn/x86_64/core2/gcd_11.asm b/mpn/x86_64/core2/gcd_11.asm
index 11aca4746..ba7e1b6d9 100644
--- a/mpn/x86_64/core2/gcd_11.asm
+++ b/mpn/x86_64/core2/gcd_11.asm
@@ -73,20 +73,21 @@ ASM_START()
 	ALIGN(16)
 PROLOGUE(mpn_gcd_11)
 	FUNC_ENTRY(2)
-	mov	v0, %rax	C
-	jmp	L(odd)		C
+	jmp	L(odd)
 
-	ALIGN(16)		C
-L(top):	cmovc	v0, u0		C u = |u - v|
-	cmovc	%r9, %rax	C v = min(u,v)
-	shr	R8(%rcx), u0	C
-	mov	%rax, v0	C
-L(odd):	sub	u0, v0		C
-	bsf	v0, %rcx	C
-	mov	u0, %r9		C
-	sub	%rax, u0	C
-	jnz	L(top)		C
+	ALIGN(16)
+L(top):	cmovc	%rdx, u0		C u = |u - v|
+	cmovc	%rax, v0		C v = min(u,v)
+	shr	R8(%rcx), u0
+L(odd):	mov	v0, %rdx
+	sub	u0, %rdx		C v - u
+	bsf	%rdx, %rcx
+	mov	u0, %rax
+	sub	v0, u0			C u - v
+	jnz	L(top)
 
-L(end):	FUNC_EXIT()
+L(end):	C rax = result
+	C rdx = 0 for the benefit of internal gcd_22 call
+	FUNC_EXIT()
 	ret
 EPILOGUE()
diff --git a/mpn/x86_64/gcd_11.asm b/mpn/x86_64/gcd_11.asm
index a10cc003e..40273e0d9 100644
--- a/mpn/x86_64/gcd_11.asm
+++ b/mpn/x86_64/gcd_11.asm
@@ -73,30 +73,29 @@ ASM_START()
 	ALIGN(16)
 PROLOGUE(mpn_gcd_11)
 	FUNC_ENTRY(2)
-	mov	u0, %rax
-
-	LEA(	ctz_table, %rdx)
+	LEA(	ctz_table, %r8)
 	jmp	L(ent)
 
-	ALIGN(16)			C              K8
-L(top):	cmovc	%rcx, %rax		C if x-y < 0   0
-	cmovc	%rdi, v0		C use x,y-x    0
-L(mid):	and	$MASK, R32(%rcx)	C	       0
-	movzbl	(%rdx,%rcx), R32(%rcx)	C	       1
-	jz	L(shift_alot)		C	       1
-	shr	R8(%rcx), %rax		C	       3
-	mov	%rax, %rdi		C	       4
-L(ent):	mov	v0, %rcx		C	       3
-	sub	%rax, %rcx		C	       4
-	sub	v0, %rax		C	       4
-	jnz	L(top)			C
-
-L(end):	mov	v0, %rax
+	ALIGN(16)
+L(top):	cmovc	%rdx, u0		C u = |u - v|
+	cmovc	%rax, v0		C v = min(u,v)
+L(mid):	and	$MASK, R32(%rdx)
+	movzbl	(%r8,%rdx), R32(%rcx)
+	jz	L(shift_alot)
+	shr	R8(%rcx), u0
+L(ent):	mov	u0, %rax
+	mov	v0, %rdx
+	sub	u0, %rdx
+	sub	v0, u0
+	jnz	L(top)
+
+L(end):	C rax = result
+	C rdx = 0 for the benefit of internal gcd_22 call
 	FUNC_EXIT()
 	ret
 
 L(shift_alot):
-	shr	$MAXSHIFT, %rax
-	mov	%rax, %rcx
+	shr	$MAXSHIFT, u0
+	mov	u0, %rdx
 	jmp	L(mid)
 EPILOGUE()